The capability layer for agents

Capabilities shouldn't live inside a prompt.

Skills are a versioned capability registry. Publish instructions, templates, and scripts once — then any agent's manifest can reference them by name, or pin an exact version for the life of a run. Every activation shows up in the run trace.

sypho skill push
↓
Skill library
├─ refund-policy@3
├─ escalation-playbook@1
└─ incident-response@2
↓
Agent reads it at run time
↓
Trace records usage

The same playbook keeps getting rebuilt.

Production teams repeat the same operational patterns across agents — support escalation, incident response, onboarding, compliance review, deployment playbooks, data workflows. Skills turn that repetition into reusable capabilities.

Instructions, resources, and workflows — one capability package.

Step-by-step guidance for specific procedures
Examples and documentation to ground decisions
Structured procedures for consistent execution
Supporting logic and validation rules
Scripts and tools run in controlled environments
Clear contracts for skill activation and results
When and how to apply the skill effectively

Published once. Referenced everywhere — pinned when it matters.

A manifest references a skill by name for always-latest, or name@version to pin it for the run's whole life, even on replay. A published version is immutable — ship a change as a new version, never a silent rewrite. sypho records when a skill is available, when it activates, and how it shapes execution.

skill.available: refund-policy
skill.activated: refund-policy
skill.resource.used: templates/refund-review.md
skill.execution.completed

Skills aren't just instructions.

Some skills include executable resources — scripts, workflow helpers — and sypho runs them inside controlled execution environments, recording the outcome in the run trace.

Skill usage should be visible, not hidden in prompts.

When a skill is activated, teams should see it in the run timeline alongside tool calls, structured-data access, logs, and results.

Capabilities that scale without duplication.

Reusable best practices

Turn repeatable procedures into shared capabilities.

Less prompt duplication

Stop copying long instructions across every agent.

Versioned, not duplicated

Publish once, reference by name or pin an exact version from any manifest.

Observable activation

See when skills are available, activated, and used.

Executable workflows

Safely combine instructions with controlled workflow helpers.

Package what your agents need to know how to do.

Request access and help shape reusable capabilities for production agents.

Request access